Gen
Use this skill to generate user-facing artifacts through the Okou CLI:
okou generate -h
okou generate is the source of truth. Always inspect the current command help when exact flags, models, styles, or providers matter.
Core Commands
okou generate image - billed image file generation; supports built-in models, image editing/reference inputs, image style registry selection, and connector guidance.
okou generate presentation - returns an Open Design resource-selection packet for an HTML presentation that the agent authors and hosts.
okou generate website - returns website authoring instructions / an Open Design packet that the agent uses to build and host a static site.
okou generate report, docs-design, poster, dashboard-design, mobile-app-design - return Open Design resource-selection packets for static HTML artifacts.
okou generate text, code, document, audio - list connector-backed options and print connector skill-invocation guidance; these do not have built-in platform pipelines unless the CLI help says otherwise.
Run okou generate <type> with no generation input to list available providers for that artifact type. Add --all when unavailable or not-yet-authorized connectors are relevant.
Generation Workflow
Identify the artifact type from the user's request.
- Use
image for raster images, edits, references, thumbnails, icons, illustrations, and visual assets.
- Use Open Design artifact commands for websites, decks, reports, posters, docs, dashboards, and mobile UI prototypes.
- Use connector-listing commands for text, code, document, or non-speech audio generation.
Discover current capability before committing.
- Run
okou generate <type> -h for flags and built-in model support.
- Run
okou generate <type> to see available providers.
- If the user asked for a connector/provider by name, run
okou generate <type> --provider <name> to get invocation guidance, then follow that provider skill.
Determine whether style discovery is needed.
- For styled images, inspect the current style registry from
okou generate image -h.
- Do not hardcode style names or style descriptions in this skill. Treat the registry printed by the CLI as the live source.
- Choose a registered style when the user's wording clearly matches a trigger, named style, or visual direction in the registry.
- To obtain style-specific prompt guidance, run
okou generate image --style <id> --prompt "<brief>" --compile, follow the returned packet, then generate with --compiled-prompt "<final prompt>".
- Use
--raw-prompt "<final prompt>" when the user explicitly wants no registry style, wants photorealism/model-native output, supplies a fully specified prompt, or no registry style is a good match.
- When no style is obvious but style materially affects the result, ask the user to choose among a few options summarized from the live registry or ask whether to proceed without a style.
Decide provider and model.
- Prefer
--provider built-in when the user wants a direct artifact, does not name a connector, and the built-in pipeline supports the request.
- Use connector guidance when the user names a provider, needs a provider-specific capability, or the requested artifact type is connector-only.
- Ask the user when the choice changes cost, latency, fidelity, licensing, account usage, or final format in a way that is not implied by the request.
- Otherwise choose a sensible default from the CLI help and proceed.
Build the prompt.
- Preserve the user's core intent, constraints, audience, brand, source materials, aspect ratio, size, format, and delivery target.
- Add operational details only when they improve generation reliability: composition, visual hierarchy, must-include/must-avoid elements, target medium, and reference handling.
- For style-guided image generation, let the selected registry style drive stylistic details through the compilation packet.
- For prompt text that is long or quote-sensitive, use a file and a safely quoted argument, or stdin when the selected prompt mode supports it.
Execute and wait for completion.
- Run the selected
okou generate <type> command.
- For commands that return an Open Design resource-selection packet, follow the packet: author the artifact, verify it locally if needed, and host static outputs with
okou host.
- For commands that return
/f/ file URLs, keep the URL and metadata for the user.
- If generation fails because of missing credits, run
okou doctor credit.
- If connector auth fails, run
okou doctor check-connector using the environment name or URL from the provider guidance.
Deliver the result.
- Give the user the generated URL or hosted artifact URL.
- Mention important parameters used: provider, model, selected style or raw prompt mode, size/aspect ratio, or site slug.
- If the output is temporary or provider-hosted with expiration, download or host a durable copy when appropriate.
Asking vs. Choosing
Ask the user before generation when:
- The prompt is too underspecified to produce a useful artifact.
- Multiple provider/style/model choices are plausible and materially different.
- The command will spend meaningful credits and the request did not imply that spend.
- The user requested brand, person, legal, medical, financial, or other high-stakes accuracy that needs source material.
- Required inputs are missing, such as source images, brand assets, dimensions, or audience.
Proceed without asking when:
- The user gave enough context for a reasonable first version.
- The built-in default clearly fits the request.
- The user asked for speed or explicitly delegated choices.
- The missing details can be safely inferred and refined after the first artifact.
Common Patterns
List current image providers:
okou generate image
Inspect current image styles and flags:
okou generate image -h
Compile a styled image prompt after selecting a live registry style, then use the packet's final prompt:
okou generate image --style "<style-id>" --prompt "<brief>" --compile
okou generate image --provider built-in --compiled-prompt "<final prompt>"
Generate an unstyled/model-native image:
okou generate image --provider built-in --raw-prompt "<prompt>"
Use connector guidance instead of built-in generation:
okou generate image --provider "<connector-name>"
Generate a static Open Design artifact:
okou generate website --prompt "<brief>"
Then follow the returned packet to build and host the artifact.
1---2name: gen3description: Use Okou generation pipelines for images, presentations, websites, reports, and designs.4metadata:5 source: https://skillmd.com/skills/vm0-ai/gen6---78# Gen910Use this skill to generate user-facing artifacts through the Okou CLI:1112```bash13okou generate -h14```1516`okou generate` is the source of truth. Always inspect the current command help when exact flags, models, styles, or providers matter.1718## Core Commands1920- `okou generate image` - billed image file generation; supports built-in models, image editing/reference inputs, image style registry selection, and connector guidance.21- `okou generate presentation` - returns an Open Design resource-selection packet for an HTML presentation that the agent authors and hosts.22- `okou generate website` - returns website authoring instructions / an Open Design packet that the agent uses to build and host a static site.23- `okou generate report`, `docs-design`, `poster`, `dashboard-design`, `mobile-app-design` - return Open Design resource-selection packets for static HTML artifacts.24- `okou generate text`, `code`, `document`, `audio` - list connector-backed options and print connector skill-invocation guidance; these do not have built-in platform pipelines unless the CLI help says otherwise.2526Run `okou generate <type>` with no generation input to list available providers for that artifact type. Add `--all` when unavailable or not-yet-authorized connectors are relevant.2728## Generation Workflow29301. Identify the artifact type from the user's request.31 - Use `image` for raster images, edits, references, thumbnails, icons, illustrations, and visual assets.32 - Use Open Design artifact commands for websites, decks, reports, posters, docs, dashboards, and mobile UI prototypes.33 - Use connector-listing commands for text, code, document, or non-speech audio generation.34352. Discover current capability before committing.36 - Run `okou generate <type> -h` for flags and built-in model support.37 - Run `okou generate <type>` to see available providers.38 - If the user asked for a connector/provider by name, run `okou generate <type> --provider <name>` to get invocation guidance, then follow that provider skill.39403. Determine whether style discovery is needed.41 - For styled images, inspect the current style registry from `okou generate image -h`.42 - Do not hardcode style names or style descriptions in this skill. Treat the registry printed by the CLI as the live source.43 - Choose a registered style when the user's wording clearly matches a trigger, named style, or visual direction in the registry.44 - To obtain style-specific prompt guidance, run `okou generate image --style <id> --prompt "<brief>" --compile`, follow the returned packet, then generate with `--compiled-prompt "<final prompt>"`.45 - Use `--raw-prompt "<final prompt>"` when the user explicitly wants no registry style, wants photorealism/model-native output, supplies a fully specified prompt, or no registry style is a good match.46 - When no style is obvious but style materially affects the result, ask the user to choose among a few options summarized from the live registry or ask whether to proceed without a style.47484. Decide provider and model.49 - Prefer `--provider built-in` when the user wants a direct artifact, does not name a connector, and the built-in pipeline supports the request.50 - Use connector guidance when the user names a provider, needs a provider-specific capability, or the requested artifact type is connector-only.51 - Ask the user when the choice changes cost, latency, fidelity, licensing, account usage, or final format in a way that is not implied by the request.52 - Otherwise choose a sensible default from the CLI help and proceed.53545. Build the prompt.55 - Preserve the user's core intent, constraints, audience, brand, source materials, aspect ratio, size, format, and delivery target.56 - Add operational details only when they improve generation reliability: composition, visual hierarchy, must-include/must-avoid elements, target medium, and reference handling.57 - For style-guided image generation, let the selected registry style drive stylistic details through the compilation packet.58 - For prompt text that is long or quote-sensitive, use a file and a safely quoted argument, or stdin when the selected prompt mode supports it.59606. Execute and wait for completion.61 - Run the selected `okou generate <type>` command.62 - For commands that return an Open Design resource-selection packet, follow the packet: author the artifact, verify it locally if needed, and host static outputs with `okou host`.63 - For commands that return `/f/` file URLs, keep the URL and metadata for the user.64 - If generation fails because of missing credits, run `okou doctor credit`.65 - If connector auth fails, run `okou doctor check-connector` using the environment name or URL from the provider guidance.66677. Deliver the result.68 - Give the user the generated URL or hosted artifact URL.69 - Mention important parameters used: provider, model, selected style or raw prompt mode, size/aspect ratio, or site slug.70 - If the output is temporary or provider-hosted with expiration, download or host a durable copy when appropriate.7172## Asking vs. Choosing7374Ask the user before generation when:7576- The prompt is too underspecified to produce a useful artifact.77- Multiple provider/style/model choices are plausible and materially different.78- The command will spend meaningful credits and the request did not imply that spend.79- The user requested brand, person, legal, medical, financial, or other high-stakes accuracy that needs source material.80- Required inputs are missing, such as source images, brand assets, dimensions, or audience.8182Proceed without asking when:8384- The user gave enough context for a reasonable first version.85- The built-in default clearly fits the request.86- The user asked for speed or explicitly delegated choices.87- The missing details can be safely inferred and refined after the first artifact.8889## Common Patterns9091List current image providers:9293```bash94okou generate image95```9697Inspect current image styles and flags:9899```bash100okou generate image -h101```102103Compile a styled image prompt after selecting a live registry style, then use the packet's final prompt:104105```bash106okou generate image --style "<style-id>" --prompt "<brief>" --compile107okou generate image --provider built-in --compiled-prompt "<final prompt>"108```109110Generate an unstyled/model-native image:111112```bash113okou generate image --provider built-in --raw-prompt "<prompt>"114```115116Use connector guidance instead of built-in generation:117118```bash119okou generate image --provider "<connector-name>"120```121122Generate a static Open Design artifact:123124```bash125okou generate website --prompt "<brief>"126```127128Then follow the returned packet to build and host the artifact.